🔍 Read the full analysis: Claude’s Training And Pirated Books: Claims Versus Proven Facts on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A June 2025 ruling found that Anthropic’s use of books to train Claude was fair use, but that downloading millions of pirated book copies for a permanent library was not. Anthropic later agreed to a $1.5 billion settlement over the downloads while denying wrongdoing; new lawsuits filed in 2026 have not been decided.
A federal judge found in June 2025 that Anthropic’s training of Claude on books was fair use, while ruling that the company’s downloads of millions of pirated book copies to build a permanent library were not. Anthropic later agreed to a $1.5 billion settlement over the downloads, and authors filed new suits in May 2026; those claims have not yet been decided.
The decision came in Bartz v. Anthropic, a case brought by three authors in 2024. In an order dated June 23, 2025, U.S. District Judge William Alsup distinguished between using books to train a model and obtaining copies unlawfully. He concluded that training Claude on books was fair use, describing the process as an effort “to turn a hard corner and create something different.” He also found that buying print books and scanning them was fair use.
The ruling treated the company’s pirated-book downloads differently. It described 196,640 books from the Books3 collection, at least five million copies from Library Genesis, and at least two million from the Pirate Library Mirror—more than seven million copies in total. Alsup found that downloading those copies for a permanent library was not fair use. The order was a pretrial ruling, not a jury verdict on damages; the case settled before a jury decided the remaining issues.
The settlement, approved in July 2026, provides $1.5 billion for owners of about 500,000 registered books downloaded from LibGen or PiLiMi, or roughly $3,000 per book before fees. Anthropic disputes wrongdoing. In the settlement process, it represented that neither dataset, nor portions of them, was in the training corpus of any commercially released large language model. The settlement requires destruction of the downloaded collections and copies, subject to legal-hold obligations, and releases specified past claims; it does not release claims about Claude’s outputs.
From download to IPO
Five years, one ruling, no trial
The piracy happened in 2021 and 2022. The only ruling came in 2025. The new lawsuits are years away from a verdict.
- Jun 2021 · Jul 2022DownloadsAt least 5 million copies from LibGen (2021), at least 2 million from PiLiMi (2022); 196,640 from Books3 earlier.
- Aug 2024Bartz v. Anthropic filedThree authors sue Anthropic, on behalf of a class of book owners.
- 23 Jun 2025Judge Alsup's orderTraining on books: fair use. Scanning bought books: fair use. Pirated library copies: not fair use, to trial.
- Sep 2025$1.5 billion settlementNo trial. Anthropic denies wrongdoing; release covers conduct up to 25 Aug 2025 only.
- 13 + 15 May 2026Opt-out suitsCruz (28 plaintiffs) and Kwon (incl. Chang-rae Lee) join Cambronne and two publisher suits.
- Jul 2026Final approvalSettlement approved. Anthropic: no opt-out deal "above the Bartz claim value".
- 24 Sep 2026Motion to dismissAnthropic: claims too late, no infringing output shown, many works already released.
- 26–27 Oct 2026NowS-1 and the headlineIPO filing cites Bartz as a risk; The Stanford Daily names Stanford authors.
- 17 Dec 2026ScheduledHearingCourt hears the motion to dismiss.
- May 2027 · Jan 2028ScheduledDiscovery, then trial settingFact discovery closes 13 May 2027; trial-setting conference 11 Jan 2028.
Sources: Bartz v. Anthropic, order of 23 June 2025; settlement FAQ; joint case statement, 27 Aug 2026; Anthropic letter, 27 Jul 2026; case management order, 14 Sep 2026; Anthropic S-1, 26 Oct 2026; The Stanford Daily, 27 Oct 2026. Scheduled dates can move.
Powered by Thorsten Meyer AI
The Ruling Separates Training From Piracy
The distinction matters because headlines describing the decision simply as a finding that Anthropic “illegally trained Claude” collapse two different questions. The judge ruled that training on books could be fair use, while finding that acquiring pirated copies for a permanent library was not. The ruling therefore does not establish that model training itself was illegal.
It also does not settle every dispute about what material influenced commercially released Claude models. The court’s pretrial order accepted the authors’ account of the pirated downloads and their use in training some models, while Anthropic’s settlement representation says the LibGen and PiLiMi datasets were not in the training corpus of any commercially released model. Those statements address different aspects of the record, and the settlement did not produce a trial verdict resolving all questions about data use.
For writers and publishers, the case sets a concrete legal and financial outcome for a large body of pirated copies, but it is not a general licence for AI companies to use copyrighted works. For AI developers, it highlights that the legal assessment can differ between training practices and how source material was obtained or stored. The new suits may test issues that the settled case did not resolve.
Products Worth Considering
HP Sprocket Photo Booth Machine Instant Color Photo Printer with 10.1 Touchscreen, LED Ring Light, 250+ Frames, 3×4 ZINK Paper, QR Code Shar
Dell UltraSharp U2725QE 27-Inch 4K UHD IPS Black Monitor
Logitech MX Brio Ultra HD 4K Webcam, Graphite
As an affiliate, we earn on qualifying purchases.
From Bartz to the New Suits
Bartz v. Anthropic began in August 2024. In June 2025, Alsup ruled on competing arguments before trial. He found the training and scanning of purchased books to be fair use, but said the use of pirated copies to create a permanent library was not. He wrote that buying a copy later would not erase liability for the earlier theft. Because the parties settled, the jury did not determine damages or decide the rest of the dispute.
The settlement covers specified past conduct through August 25, 2025. It provides payment to owners of registered books found in the downloaded collections and requires destruction of those copies. In May 2026, authors filed separate lawsuits raising new claims. A Stanford Daily report identified several Stanford-associated writers whose books it said were among the pirated copies, including Tobias Wolff, Abraham Verghese, Chang-rae Lee, Daniel Mason, Hayagreeva Rao and Myra Strober. The report said ten Wolff works were affected; that book list has not been independently verified here.
The report also described a May 2026 ruling that found Anthropic pirated books. The cited order, however, is from the June 2025 Bartz case, not a May 2026 decision. Chang-rae Lee is listed as a plaintiff in one of the new suits, according to the source account. The available filings reviewed in that account did not list Wolff among plaintiffs in five related cases as of August 2026; whether he appears in another case or a later amendment was not established.
““Buying a copy later will not absolve it of liability for the theft.””
— Judge William Alsup, discussing the pirated copies
Questions the Settlement Left Open
The new authors’ lawsuits remain unresolved. No court has ruled on their claims, and a hearing on Anthropic’s motion to dismiss is scheduled for December 17, 2026. The available material does not establish which claims will proceed after that hearing or whether the parties will settle.
The precise relationship between the pirated collections and training of particular Claude models remains contested in the public account. The 2025 order addressed the parties’ pretrial record, while Anthropic’s settlement representation concerned the training corpus of commercially released models. The settlement ended the original case without a jury deciding the remaining factual issues. The Stanford Daily’s specific list of affected books also has not been checked against the settlement database in the source account.
Claims about other language in Anthropic’s public filing should be read separately from the copyright case. The source account says the company’s S-1 described model behavior “resembling blackmail,” concerns about possible AI sentience and existential risks raised by others; it does not say Claude is sentient or that it could blackmail the public. A staged safety test cited in coverage is not proof that the behavior occurred in ordinary use.
December Hearing Will Test New Claims
The next stated court milestone is the December 17, 2026 hearing on Anthropic’s motion to dismiss the new author claims. The judge’s decision could determine which parts of the litigation move forward, but the outcome is not known. Any later proceedings would need to address allegations that were not resolved by the Bartz settlement.
Separately, the settlement’s payment schedule calls for installments, including $450 million by September 25, 2026, and a further $450 million by September 27, 2027, according to the terms summarized in the source material. The record provided does not confirm the status of each payment. Readers should distinguish those settlement obligations from the pending lawsuits: the settlement addresses defined past claims, while the new cases seek rulings on claims still before the court.
Key Questions
Did a judge rule that Claude was illegally trained on pirated books?
Not in those terms. In June 2025, Judge Alsup found training on books to be fair use, but ruled that downloading pirated copies to build a permanent library was not. The case settled before a jury decided the remaining issues.
What does Anthropic’s $1.5 billion settlement cover?
The settlement provides payments to owners of about 500,000 registered books downloaded from LibGen or PiLiMi and requires destruction of those downloaded copies, subject to legal-hold obligations. It covers specified past conduct and does not release claims about Claude’s outputs.
Have the lawsuits filed in 2026 been decided?
No. The new author claims remain pending. A hearing on Anthropic’s motion to dismiss is set for December 17, 2026; no ruling on those claims is reported in the source material.
Does Anthropic say the pirated collections trained released Claude models?
No. In its settlement representation, Anthropic said neither LibGen nor PiLiMi, nor portions of those datasets, was in the training corpus of any commercially released model. That is the company’s stated position in the settlement, not a jury finding after trial.
Source: ThorstenMeyerAI.com
